CN110853018A - Computer vision-based vibration table fatigue crack online detection system and detection method - Google Patents

Abstract

The invention discloses an on-line detection system for vibration table fatigue cracks based on computer vision, which comprises an image acquisition device, a bracket, a multi-channel data collector, a processor, a display and a crack detection module. The image/video information is collected in the areas where cracks are prone to occur on the surface of the parts. The multi-channel data collector aggregates the multi-channel data into one signal and transmits it to the processor. The crack detection module extracts the comparison frame, eliminates vibration displacement, features screening, and shape screening. , crack verification, calculation of actual crack length, statistics of crack information and auxiliary reminders, accumulation of positive and negative samples, and parameter adaptive adjustment to perform online detection of fatigue cracks, and output the crack information results through the display. The invention can self-adaptively eliminate the displacement generated by the vibration table, and robustly obtain information such as the position, length and shape of the crack of the parts, so as to realize the automatic detection of the crack.

Description

An online detection system and detection of vibration table fatigue cracks based on computer vision method

technical field

The invention relates to the field of video analysis and image processing, in particular to an on-line detection system and detection method for vibration table fatigue cracks based on computer vision.

Background technique

The fatigue fracture result of metal parts is an important indicator for evaluating the quality of metal parts. Accurate evaluation of the quality of metal parts can effectively reduce major safety accidents and economic losses caused by quality reasons during use. However, the existing traditional fatigue crack detection methods are mainly visual inspection methods, which are difficult to read during the test, need to be stopped for measurement, low measurement accuracy, and are easily affected by human factors. If you want to achieve real-time online observation, you need to use More manpower, very time-consuming and labor-intensive. Therefore, how to accurately and reliably detect the initiation and propagation of fatigue cracks in real time is crucial to establishing accurate and reliable fatigue tests, saving manpower and improving work efficiency.

At present, the online crack detection methods mainly include ultrasonic infrared, eddy current, acoustic emission and computer vision-based detection methods. The ultrasonic infrared method uses the interaction between ultrasonic excitation energy and defects to generate temperature changes, and then collects temperature field information through a thermal imager to detect crack information, as proposed in the document "Application of Ultrasonic Infrared Thermal Imaging Technology in Metal Crack Detection" The disadvantage is that due to the propagation of heat, the precise shape of the crack cannot be obtained, and it is difficult to achieve dynamic real-time detection; eddy current is based on the eddy current phenomenon of electromagnetics, and a high-frequency signal is applied to the tested specimen. When there are defects on the surface, the eddy current density changes locally, and then the surface temperature distribution changes, and then the temperature field information is collected by the thermal imager to detect the crack information, such as the method proposed in the literature "Eddy Current Excitation Thermal Imaging Metal Weld Crack Detection Method" , but in actual testing, the electromagnetic properties, thermal properties and other parameters of materials vary widely, eddy current testing is easily affected by external magnetic fields, only ferromagnetic materials can be detected, and the detection depth is only 2-3mm; the acoustic emission method uses the acoustic emission acquisition device , Microscopic image acquisition device and fatigue crack state real-time monitoring device, process the received signal, generate acoustic emission characteristic parameters, graphics and crack size data of the fatigue crack change process, and then realize online dynamic and real-time detection of fatigue crack initiation and propagation, Such as the method proposed in patent CN201310467034.7; the crack detection method based on computer vision detects crack information through design features, such as the method proposed in invention patent CN201510916229.4, the existing method is only for static metal parts. However, there is no suitable method and system for the metal parts that vibrate rapidly on the vibration table in the fatigue fracture test. In view of this, this patent proposes an online detection system and detection method for vibration table fatigue cracks based on computer vision.

SUMMARY OF THE INVENTION

In order to overcome the defects of the prior art, the present invention provides an on-line detection system and detection method for fatigue cracks in the whole process from generation to expansion to fracture based on video images for the shaking table environment, which can adaptively eliminate the displacement generated by the shaking table, Real-time detection of the time, position, length and shape of the cracks on the vibration table, and automatically send the information to the relevant personnel to control the vibration table to stop.

The invention provides an on-line detection system for vibration table fatigue cracks based on computer vision, which includes an image acquisition device, a bracket, a multi-channel data acquisition device, a processor, a display and a crack detection module, and the image acquisition device is fixed on the bracket. and by adjusting the position of the bracket to collect image/video information on the area where the surface of the parts is prone to cracks; The processor performs processing, the display is connected to the processor through a video cable, the crack detection module is installed in the processor, performs online detection of fatigue cracks, and outputs the detection results through the display The image acquisition device includes a plurality of cameras for collecting image/video data information at different positions on the surface of the parts, and the number of the cameras is determined by the specific positions where cracks are prone to occur on the surface of the parts; the brackets are used for fixing multiple The multi-channel data collector is used to synchronously collect the image/video data collected by multiple cameras together; the processor is used to realize the collection and storage of data, and provide a carrier for the crack detection module ; The crack detection module includes an image acquisition module, a human-computer interaction module, a data storage module and a crack detection algorithm module, and the image acquisition module is used to collect image/video data transmitted by the multi-channel data collector; the human The computer interaction module is used to interactively set various parameter information, and to determine the initial position of crack detection and the correctness of detection; the data storage module is used to record image data and detection logs during the detection process, and the crack detection algorithm module is used to Analyze the image data and obtain information such as the position, length and shape of the crack through the designed crack detection algorithm.

Preferably, the support includes a retractable rectangular frame, a magnetic base and a suspension device, the rectangular frame includes four uprights and a top surface assembly, the magnetic base is installed at the bottom of each upright, and the uprights pass through The adjustment locking device has a telescopic function; the top surface assembly includes at least 3 beams and 2 longitudinal beams, the beams are cantilever beams, and the longitudinal beams have a telescopic function by adjusting the locking device. The beam can slide forward and backward along the longitudinal beam, and a plurality of suspension devices are arranged on the beam for fixing the camera, which can slide along the beam and have a telescopic function; the suspension device includes a telescopic rod and a telescopic function. The gimbal is used for fixing the camera and adjusting the orientation of the camera. The interior of the retractable rod is hollow and used for arranging the camera connection line.

The present invention also provides an online detection method for vibration table fatigue cracks based on computer vision, wherein the crack detection algorithm module realizes functions through the following steps:

Step 1, comparison frame extraction: at the beginning of detection, capture a frame without cracks as a comparison frame;

Step 2, vibration displacement elimination: by finding the most similar frame to the comparison frame in the consecutive n frames of images and fine-tuning the target frame position by shifting the target frame, thereby eliminating the displacement of the parts due to vibration;

Step 3, feature screening: by screening the three features of the difference region, color region and position region of the target frame, the region satisfying the three features at the same time is obtained as the abnormal region;

Step 4, morphological screening: perform contour tracing on the abnormal area and perform morphological screening on the obtained contour, and select the contour whose contour length is greater than a specific value, a bar shape and a similar direction as the abnormal contour;

Step 5, crack verification: calculate the difference between the pixel gray values on both sides of the abnormal contour, remove the abnormal contour whose difference average value is greater than the threshold T _T , obtain the crack contour, and record the location and time information of the crack generation at this time;

Step 6, calculate the actual length of the crack: mark the dots representing the standard distance length on the surface of the part, and use contour tracing to calculate the pixel length between the two marked dots and the pixel length of the detected crack outline, according to the surface of the part. The geometric shape calculates the transformation relationship between the actual length between the two marked dots and the pixel length, and then calculates the actual length of the crack outline according to the transformation relationship;

Step 7: Statistical crack information and auxiliary reminders: Count the crack position, length change, crack image and time information during the entire process from the occurrence to the propagation of the crack until reaching the maximum allowable crack length, and send the information to the relevant monitoring personnel remotely. Decide whether to record relevant information or control the shaking table to stop movement according to the crack detection result;

Step 8, accumulation of positive and negative samples and adaptive adjustment of parameters: the operator can periodically replay the historical records saved in the database through the data storage module, and support manual measurement of the crack pixel length along the crack direction, through the actual distance between the two marked dots. The conversion relationship between the length and the pixel length calculates the actual length of the crack, realizes the verification of historical data and determines the correct positive samples and negative samples with errors calculated by the crack detection algorithm, and saves the positive and negative samples of crack detection confirmed by manual verification to the database. It is used to train and iteratively adjust each threshold in the algorithm module, and continuously optimize the recognition accuracy of the algorithm, so that the threshold setting can achieve the best detection effect.

Preferably, the step 2 specifically includes the following steps:

Step 2.1, rough vibration elimination: every certain time, the camera continuously captures multiple frames of images, makes frame differences between the multiple frames of images and the comparison frames one by one, and takes the frame with the smallest difference as the target frame;

Step 2.2, fine vibration removal: keep the comparison frame unchanged, make a certain range of translation of the target frame in the positive and negative directions of the x-axis and make a frame difference with the target frame, and take the smallest frame difference result as the offset distance on the x-axis , in the same way, calculate the offset distance in the y-axis direction, and move the target frame in the x- and y-axis directions respectively by the corresponding offset distance to eliminate the displacement of the components.

Preferably, the step 3 specifically includes the following steps:

Step 3.1, obtain the difference area: make a frame difference between the contrast frame and the target frame that eliminates the displacement, and take the area where the gray value of the frame difference result is greater than the threshold _TB , which conforms to the generation characteristics of the crack area from scratch, and the result is area containing crack information;

Step 3.2, obtaining a darker area: in the target frame, select an area whose gray value is less than the threshold value T _A , which conforms to the darker color characteristics of the crack area, and obtains a darker area;

Step 3.3, obtaining the component area: perform a binarization operation with a threshold value of T _w on the comparison frame, obtain the contour of the binarization result, select the area surrounded by the largest contour as the component area, and make a histogram for the component area Stretch and select an area whose gray value is less than the threshold _TD in the component area to obtain the final component area, which conforms to the location characteristics of cracks;

Step 3.4, synthesizing multiple features: selecting an area that satisfies the three conditions of steps 3.1, 3.2, and 3.3 at the same time as an abnormal area.

Preferably, the step 4 specifically includes the following steps:

Step 4.1, contour length screening: further screen the contour lines in the abnormal area, and select the contour whose contour length is greater than the threshold _TL ;

Step 4.2, outline aspect ratio screening: select the outline whose aspect ratio of the smallest circumscribed rectangle is greater than the threshold _TR ;

Step 4.3, direction similarity screening: select the contour whose contour line direction is greater than the proportional threshold _TP .

Compared with the prior art, the present invention has the following beneficial effects:

1. The present invention can realize real-time online detection of fatigue cracks in the shaking table scene, saving a lot of manpower and costs;

2. The invention adopts the method based on computer vision to realize crack detection, and has the advantages of intuitive positioning, high sensitivity, strong adaptability, and convenient layout;

3. The crack detection method proposed by the present invention integrates the characteristics of various crack regions and can adaptively eliminate the displacement generated by the shaking table, and has stronger robustness.

Description of drawings

Fig. 1 is the structural representation of the vibration table fatigue crack online detection system based on computer vision of the present invention;

2 is a schematic diagram of a bracket structure of a computer vision-based on-line detection system for vibration table fatigue cracks in an embodiment of the present invention;

Fig. 3 is the human-computer interaction module operation interface of the vibration table fatigue crack online detection system based on computer vision of the present invention;

Fig. 4 is the data storage module operation interface of the vibration table fatigue crack online detection system based on computer vision of the present invention; And

FIG. 5 is a schematic flowchart of the on-line detection method for vibration table fatigue cracks based on computer vision according to the present invention.

Detailed ways

Exemplary embodiments, features and performance aspects of the present invention will be described in detail below with reference to the accompanying drawings. The same reference numbers in the figures denote elements that have the same or similar functions. While various aspects of the embodiments are shown in the drawings, the drawings are not necessarily drawn to scale unless otherwise indicated.

As shown in Figure 1, an online fatigue crack detection system for a shaking table based on computer vision includes an image acquisition device, a bracket, a multi-channel data acquisition device, a processor, a display, and a crack detection module.

When testing, first place the bracket near the component, and each camera in the image acquisition device is fixed on the corresponding suspension device on the bracket by screws, facing the surface of the component to be inspected, that is, the area prone to cracks, to collect image/video information ;Multiple cameras are connected to the multi-channel data collector through the data cable, and the multi-channel data collector synchronizes and aggregates the collected multi-channel image/video data into one signal and transmits it to the processor for processing; the display and the processor are connected through the video cable Connected, the multi-channel data collector, the processor and the display are respectively connected to the power supply. After the crack detection module is activated, the image acquisition equipment starts to work. The user needs to first adjust the locking device of the bracket and the retractable rod and gimbal on the suspension device according to the surface image of the part collected on the interface, so that each camera is aligned. In areas prone to cracks, adjust the focus of the camera to make the picture clear; further rotate the knob of the magnetic base to fix the bracket on the vibration table. After the working field of view of the camera is determined, the crack detection module officially starts to detect. Through the designed crack detection algorithm, the position, length and shape of the crack can be obtained. The results are output through the display, and the relevant personnel are notified through voice prompts and emails.

The image acquisition device includes a plurality of cameras, which are used to acquire image/video data of image/video data information at different positions on the surface of the part. In this example, Daheng MER-051-120U3M-L camera is used, the resolution is 808(H)*608(V), the frame rate is 120fps, the data interface is USB3.0, and the camera lens adopts Daheng A3Z3112CS-MPIR zoom Lens, the focal length range is 3.1-8mm, and the focusing method is manual adjustment.

The bracket includes a retractable rectangular frame, a magnetic base and a suspension device, which is mainly used to flexibly arrange multiple cameras to adapt to the crack detection positions of parts of different shapes and sizes. As shown in FIG. 2 , in this embodiment, the rectangular frame includes four uprights and a top surface assembly, and two sections of hollow thick steel pipes 3 are connected by adjusting the locks 2 to form uprights with adjustable heights, and a magnetic force is installed at the bottom of each upright. Seat 1, used to fix the whole bracket. The top surface assembly includes three beams 6 and two longitudinal beams. The three-way connector 4 is used to connect three vertical columns, beams and longitudinal beams in a vertical direction. The beam is a cantilever beam, used to install suspension devices, with lock The sliding bearing 5 of the clamping device can slide freely on the beam; the longitudinal beam has a telescopic function by adjusting the lock 11 to adjust the length of the connecting rod, and the beam in the middle can be connected in three directions through the sliding bearing with the locking device 10Slide back and forth along the stringer. There are 6 suspension devices on the beam to fix the camera. The sliding bearing 5 with locking device and the adjusting lock 8 can slide along the beam and have a telescopic function. The suspension device includes a telescopic rod and a gimbal 9 arranged at the end. The telescopic rod is composed of two thin hollow steel pipes 7 and an adjustment lock 8 connecting the two thin hollow steel pipes. The interior is hollow and used for arranging camera connections. Line; the gimbal 9 is composed of a gimbal head and a nine-hole fixing plate, which is used to fix the camera and adjust the orientation of the camera.

The multi-channel data collector is used to synchronously collect the image/video data collected by the multi-channel cameras. In this embodiment, this function is implemented through an industrial-grade 20-channel HUB, and data is transmitted to the processor through a USB data line.

The processor is mainly used to realize data collection, storage and crack detection methods, and provides a carrier for the crack detection module. In this embodiment, the processor is a personal PC.

The display is mainly used to display the part image and inspection data information. In this embodiment, the display is a 27-inch liquid crystal display.

The crack detection module is installed in the processor and is used for online detection of fatigue cracks, including an image acquisition module, a human-computer interaction module, a data storage module, and a crack detection algorithm module.

The image acquisition module is mainly used to collect the image/video data transmitted by the multi-channel data collector; in this embodiment, the video acquisition function provided by Opencv is used to directly collect images and videos;

The human-computer interaction module is mainly used to preset parameters and adjust each camera to suit the working conditions. When entering the software system, the human-computer interaction module pops up, and after completing the setting, data acquisition and crack detection can be started. As shown in FIG. 3 , in this embodiment, the parameters need to set the crack detection interval time TG and the part number. If it is set to perform a crack detection task every 2 minutes, the part number is s00101. After initialization, the s00101 folder will be generated locally on the computer to store all image records and inspection logs of the part. This module supports adjusting the focal length of the camera and adjusting the bracket for each camera to capture the part image with a clear picture and a suitable angle, and also supports setting the detection range for the camera picture, that is, drawing polygons in the camera picture to limit the detection range to the camera picture. Some specific parts of the components can reduce the interference caused by complex backgrounds and improve the crack detection accuracy.

The data storage module is mainly used to record the image data and detection log in the detection process. The image data is used for the operator to play back and confirm whether the information is recorded correctly; the inspection log includes information such as the time when the crack appeared, the position, length and shape of the crack. All data is saved to a database for easy operator verification. As shown in Figure 4, in this embodiment, the image data saves a frame of image every TG minutes as image data and records it in a folder named after the part number, and the frame image is accompanied by the crack length result and the corresponding detection time, The inspection log includes crack length and corresponding time information, such as: inspection time is 3 hours and 50 minutes, and crack length is 4 mm. Relevant personnel can quickly browse the overall inspection status of the component through the inspection log.

The crack detection algorithm module is mainly used to analyze the image data and obtain information such as the position, length and shape of the crack through the designed crack detection algorithm. As shown in Figure 5, the crack detection algorithm module realizes the function through the following steps:

Step 1, comparison frame extraction: at the first TG at the beginning of detection, a frame without cracks is captured as a comparison frame PC _. In this embodiment, the camera continuously captures n frames of images, selects the clearest frame among the n frames of images as the comparison frame, and adopts the image definition method based on the Tenengrad gradient function, that is, the Sobel algorithm is used to extract the horizontal and vertical directions respectively. Gradient value, the expression is as follows:

D(f)=∑ _y ∑ _x |G(x,y)|(G(x,y)>T) (1)

G(x,y) is as follows:

Step 2, vibration displacement elimination: by finding the most similar frame to the comparison frame in the consecutive n frames of images and shifting the target frame to fine-tune the position of the target frame, so as to eliminate the displacement caused by vibration.

Further, step 2 specifically includes the following steps:

Step 2.1, coarse vibration elimination: every time t, the camera continuously captures n frames of images, makes frame differences between the continuous n frames of images and the comparison frames one by one, and takes the frame with the smallest difference as the target frame P _A . In this embodiment, the interval time t is equal to the initially set detection interval time TG, and n=40 frames of continuously captured images are taken.

Step 2.2, fine vibration removal: keep the comparison frame PC unchanged, make _{a certain range of translation of the target frame P A in} the positive and negative directions of the x-axis, and make a frame difference with the target frame, and take the smallest frame difference result as the x-axis. The offset distance D _x , and the offset distance D _y in the y-axis direction is calculated in the same way. The component displacement can be finely eliminated by moving the target frame P _A in the x-axis and y-axis directions by D _x and D _y respectively. In this embodiment, the moving range of the target frame on the x-axis and the y-axis is (-50, +50) pixels, and the optimal offset distance is calculated within this range.

Step 3, feature screening, by screening the target frame for three features: difference area, color area, and position area, an area satisfying the three characteristics at the same time is obtained as an abnormal area, that is, a crack area may exist.

Further, step 3 specifically includes the following steps:

Step 3.1, obtain the difference area: make a frame difference between the comparison frame PC _C and the target frame _{P A} for which the displacement is eliminated, and take the area where the frame difference result is greater than the threshold value TB, which conforms to the generation characteristics of the crack area from scratch, and the obtained result is is the area _PDF containing crack information. In this example, the threshold T _B =40.

Step 3.2, obtaining a darker area: in the target frame PA, select an area whose gray value is less than the threshold value _TA , which corresponds to _a darker area and conforms to the darker color characteristics of the crack area, and obtains a darker area P _DA . In this embodiment, a binarization method is used to obtain a darker area, and the binarization threshold is T _A =100.

Step 3.3, get the component area: perform the _binarization operation on the comparison frame PC, the threshold is Tw, obtain the outline of the binarization result and select the largest outline as the component area, and stretch the component area by histogram. From the component area, the area with the gray value less than the threshold _TD is taken out, that is, the original dark area on the component is removed to obtain the final component area P _W , which conforms to the possible location characteristics of cracks. In this embodiment, the component area limits the crack detection position within the component area, and removes the texture in the background outside the component area (eg, vibration table) to cause misidentification, and the threshold _Tw =140. The component area is stretched by histogram, and the original dark area of the component is obtained by the _binarization method. Holes and stains cause misidentification.

Step 3.4, synthesizing multiple features: select the area that satisfies the features of steps 3.1, 3.2, and 3.3 at the same time, that is, select the area that satisfies the three features of large enough difference, dark enough color and three features in the component area as the area that may have cracks . In this embodiment, the result graph PDF , _PDA , and _PW obtained in steps 3.1, 3.2, and 3.3 are combined with operations to _define an area that satisfies the three characteristics at the same time.

Step 4, morphological screening, contour tracing is performed on the abnormal area, morphological screening is performed on the obtained contour, and a contour with a sufficiently large, bar-shaped and similar direction is selected as the abnormal contour _CP , that is, the possible crack contour.

Further, step 4 specifically includes the following steps:

Step 4.1, contour length screening: further screen the contour lines in the abnormal area, and select the contour whose contour length is greater than the threshold value _TL . In this embodiment, the contour length threshold is taken as T _L =40, that is, the contour lines whose contour length is less than 40 are removed;

Step 4.2, outline aspect ratio screening: take the outline whose aspect ratio of the smallest circumscribed rectangle is greater than the threshold _TR . In this example, the threshold of the aspect ratio of the minimum circumscribed rectangle of the contour line is taken as T _R =3.5, that is, the contour line with the length to width ratio of the minimum circumscribed rectangle less than 3.5 is removed, which conforms to the crack contour and presents strip-like features.

Step 4.3, direction similarity screening: PCA principal component analysis is used to calculate the main direction of the contour line, and the points on the contour line are encoded in the direction. Assuming that the number of contour line pixel points is m, if the absolute value of the difference between the contour line direction and the main direction If it is less than 10 degrees, the direction of the pixel is considered to be similar to the main direction. If the number of similar pixels in the total pixels is greater than the threshold _TP , the overall direction of the outline is considered to be similar. In this embodiment, angle=10°, T _P =40%, and the determination expression for the similarity of the contour lines is as follows:

Among them, a _p is the main direction of the contour line, and a _i is the direction of the ith point on the contour line.

Among them, m is the total number of pixel points of the contour line, and r is the result of the similarity judgment of the direction of the contour line.

Step 5, crack verification: Calculate the difference between the pixel gray values on both sides of the possible crack contour, and remove the possible crack contour whose mean value is greater than the threshold TT, that is, remove the difference in light on both sides of the ridge caused by the existence of a ridge structure on the surface of the part. The resulting false crack profile, the location and time information of the crack generation are recorded at this time. In this example, the threshold T _T =50, that is, to remove possible crack contours whose mean value of the difference between the grayscale values of the pixels on both sides is greater than 50.

Step 6: Calculate the actual length of the crack: Mark the dots representing the standard distance length on the surface of the part, and use contour tracing to calculate the pixel length between the two marked dots and the detected crack contour respectively. According to the surface geometry of the part The shape calculates the transformation relationship between the actual length between the two marked dots and the pixel length, and then calculates the actual length of the crack outline according to the transformation relationship. In this embodiment, two marked dots with a distance of 15 mm are set. If the two marked dots are k pixels apart in the image, the conversion relationship is c=15/k. If the detected crack outline length is p pixels, then the actual crack length l=p*c, the unit is mm.

Step 7: Statistical crack information and auxiliary reminders: Count the crack position, length change, crack image and time information during the whole process from the occurrence to the propagation of the crack until reaching the maximum allowable crack length, and output the results through the display, and send the information to the remote. For the relevant monitoring personnel, the monitoring personnel decide whether to record relevant information according to the crack detection results or control the shaking table to stop moving.

Step 8, positive and negative sample accumulation and parameter adaptive adjustment: the operator can periodically replay the historical records saved in the database, and support manual measurement of the crack pixel length along the direction of the crack, through the actual length and pixel length between the two marked dots The conversion relationship between the two measures the actual length of the crack, so as to realize the positive samples with accurate calculation results of the accumulation algorithm and the negative samples with errors in the calculation. In this example, the operator can calculate the actual length of the crack by clicking on the starting position of the crack and attach the actual length to the image to display on the software interface. The specific test results are shown in Table 1. Save the positive and negative samples of crack detection confirmed by manual verification to the database for training and iteratively adjust the eight thresholds _TB , _TA , _Tw , _TD , _TL , _TR , _TP , _TT in the algorithm, Continuously optimize the recognition accuracy of the algorithm, so that the threshold setting can achieve the best detection effect.

Table 1 Test data results of actual length and measured length

serial number Actual length (mm) Calculated length (mm) serial number Actual length (mm) Calculated length (mm) 1 14.10 13.35 12 2.20 2.12 2 20.20 18.83 13 9.80 9.45 3 16.80 15.56 14 10.00 9.78 4 7.50 7.44 15 10.20 9.89 5 2.00 2.11 16 15.30 14.92 6 4.80 4.99 17 14.80 14.66 7 10.60 10.30 18 20.10 20.06 8 7.30 7.16 19 18.60 17.94 9 8.90 8.66 20 12.70 12.51 10 9.00 8.72 twenty one 11.69 11.38 11 2.50 2.29

The computer vision-based on-line detection system and detection method for vibration table fatigue cracks provided by the present invention can adaptively eliminate the displacement generated by the vibration table, and robustly obtain the cracks of the vibration table components by aiming at the crack design characteristics in the vibration table environment. Information such as position, length, and shape can be used to automatically detect cracks.

Finally, it should be noted that the above-mentioned embodiments are only used to illustrate the technical solutions of the present invention, but not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that : it can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements to some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the technical solutions of the embodiments of the present invention range.

Claims (6)

1. a vibration table fatigue crack online detection system based on computer vision, is characterized in that, it comprises image acquisition equipment, support, multi-channel data collector, processor, display and crack detection module, and described image acquisition equipment is fixed on On the bracket and by adjusting the position of the bracket, image/video information is collected on the surface of the parts where cracks are likely to occur; the multi-channel data collector synchronizes and aggregates the collected multi-channel image/video data into one channel The signal is sent to the processor for processing, the display is connected to the processor through a video cable, and the crack detection module is installed in the processor to perform on-line detection of fatigue cracks, and the detection results are passed through the the display output; The image acquisition device includes a plurality of cameras for collecting image/video data information at different positions on the surface of the part, and the number of the cameras is determined by the specific positions on the surface of the part that are prone to cracks; the bracket is used for fixing a plurality of the cameras; The multi-channel data collector is used to synchronously collect image/video data collected by multiple cameras together; The processor is used to realize data collection and storage, and provide a carrier for the crack detection module; The crack detection module includes an image acquisition module, a human-computer interaction module, a data storage module, and a crack detection algorithm module, and the image acquisition module is used to collect image/video data transmitted by the multi-channel data collector; the human-computer The interactive module is used to interactively set various parameter information, and to determine the initial position of crack detection and the correctness of detection; the data storage module is used to record image data and detection logs in the detection process, and the crack detection algorithm module is used to analyze The image data and the crack detection algorithm are designed to obtain information such as the position, length and shape of the crack. 2. The computer vision-based on-line detection system for vibration table fatigue cracks according to claim 1, wherein the support comprises a retractable rectangular frame, a magnetic base and a suspension device, and the rectangular frame comprises four uprights and The top surface assembly, the bottom of each column is installed with the magnetic base, and the column has a telescopic function by adjusting the locking device; the top surface component includes at least 3 beams and 2 longitudinal beams, the beams It is a cantilever beam, the longitudinal beam has a telescopic function by adjusting the locking device, the transverse beam in the middle can slide forward and backward along the longitudinal beam, and a plurality of suspension devices are arranged on the transverse beam to fix the camera, It can slide along the beam and has a telescopic function; the suspension device includes a telescopic rod and a gimbal, the gimbal is used to fix the camera and adjust the orientation of the camera, and the telescopic rod is arranged inside It is hollow for arranging the camera cable. 3. the detection method of the vibration table fatigue crack online detection system based on computer vision according to claim 1 and 2, is characterized in that, described crack detection algorithm module realizes function through the following steps: Step 1, comparison frame extraction: at the beginning of detection, capture a frame without cracks as a comparison frame; Step 2, vibration displacement elimination: by finding the most similar frame to the comparison frame in the consecutive n frames of images and fine-tuning the target frame position by shifting the target frame, thereby eliminating the displacement of the parts due to vibration; Step 3, feature screening: by screening the three features of the difference region, color region and position region of the target frame, the region satisfying the three features at the same time is obtained as the abnormal region; Step 4, morphological screening: perform contour tracing on the abnormal area and perform morphological screening on the obtained contour, and select the contour whose contour length is greater than a specific value, a bar shape and a similar direction as the abnormal contour; Step 5, crack verification: calculate the difference between the pixel gray values on both sides of the abnormal contour, remove the abnormal contour whose difference average value is greater than the threshold T _T , obtain the crack contour, and record the location and time information of the crack generation at this time; Step 6, calculate the actual length of the crack: mark the dots representing the standard distance length on the surface of the part, and use contour tracing to calculate the pixel length between the two marked dots and the pixel length of the detected crack outline, according to the surface of the part. The geometric shape calculates the transformation relationship between the actual length between the two marked dots and the pixel length, and then calculates the actual length of the crack outline according to the transformation relationship; Step 7: Statistical crack information and auxiliary reminders: Count the crack position, length change, crack image and time information during the entire process from the occurrence to the propagation of the crack until reaching the maximum allowable crack length, and send the information to the relevant monitoring personnel remotely. Decide whether to record relevant information or control the shaking table to stop movement according to the crack detection result; Step 8, accumulation of positive and negative samples and adaptive adjustment of parameters: the operator can periodically replay the historical records saved in the database through the data storage module, and support manual measurement of the crack pixel length along the crack direction, through the actual distance between the two marked dots. The conversion relationship between the length and the pixel length calculates the actual length of the crack, realizes the verification of historical data and determines the correct positive samples and negative samples with errors calculated by the crack detection algorithm, and saves the positive and negative samples of crack detection confirmed by manual verification to the database. It is used to train and iteratively adjust each threshold in the algorithm module, and continuously optimize the recognition accuracy of the algorithm, so that the threshold setting can achieve the best detection effect. 4. The computer vision-based on-line detection method for vibration table fatigue cracks according to claim 3, wherein the step 2 specifically comprises the following steps: Step 2.1, rough vibration elimination: every certain time, the camera continuously captures multiple frames of images, makes frame differences between the multiple frames of images and the comparison frames one by one, and takes the frame with the smallest difference as the target frame; Step 2.2, fine vibration removal: keep the comparison frame unchanged, make a certain range of translation of the target frame in the positive and negative directions of the x-axis and make a frame difference with the target frame, and take the smallest frame difference result as the offset distance on the x-axis , in the same way, calculate the offset distance in the y-axis direction, and move the target frame in the x- and y-axis directions respectively by the corresponding offset distance to eliminate the displacement of the components. 5. the on-line detection method of vibration table fatigue crack based on computer vision according to claim 3, is characterized in that, described step 3 specifically comprises the following steps: Step 3.1, obtain the difference area: make a frame difference between the contrast frame and the target frame that eliminates the displacement, and take the area where the gray value of the frame difference result is greater than the threshold _TB , which conforms to the generation characteristics of the crack area from scratch, and the result is area containing crack information; Step 3.2, obtaining a darker area: in the target frame, select an area whose gray value is less than the threshold value T _A , which conforms to the darker color characteristics of the crack area, and obtains a darker area; Step 3.3, obtaining the component area: perform a binarization operation with a threshold value of T _w on the comparison frame, obtain the contour of the binarization result, select the area surrounded by the largest contour as the component area, and make a histogram for the component area Stretch and select an area whose gray value is less than the threshold _TD in the component area to obtain the final component area, which conforms to the location characteristics of cracks; Step 3.4, synthesizing multiple features: selecting an area that satisfies the three conditions of steps 3.1, 3.2, and 3.3 at the same time as an abnormal area. 6. The on-line detection method of vibration table fatigue crack based on computer vision according to claim 3, is characterized in that, described step 4 specifically comprises the following steps: Step 4.1, contour length screening: further screen the contour lines in the abnormal area, and select the contour whose contour length is greater than the threshold _TL ; Step 4.2, outline aspect ratio screening: select the outline whose aspect ratio of the smallest circumscribed rectangle is greater than the threshold _TR ; Step 4.3, direction similarity screening: select the contour whose contour line direction is greater than the proportional threshold _TP .

Images